The AI you can keep
on your shelf.
Lexindex indexes your case archive, incoming scans, and the Polish legal corpus on an edge server controlled by your office. Standard inference runs in the EU. Enterprise keeps inference on-prem when policy requires it.
The system shows the source document, page, and cited passage behind the answer.
Two zones.
One narrow channel.
Your archive stays on the edge. Generative work crosses one narrow, EU-only channel. Enterprise keeps inference on-prem when policy requires zero-hop.
Ingest
Office documents and Polish law are indexed locally. Updates are incremental after that.
Query
Search the local index. Most answers never leave the office.
Inference
Only the minimum tokens transit when generative help is needed.
Return
The result returns to your office; nothing is retained centrally.
Audit
Every hop is logged on your edge server. Exportable.
What is sent.
What stays.
What is never sent.
- Full case files (PDF, DOC, scans)
- SQL records · debtor / creditor data
- The local index & embeddings
- Encryption keys for the edge store
- User identities & the audit log
- Minimum prompt tokens (the question + cited snippets)
- Page images for OCR, only when explicitly invoked
- The inference result, on the way back
Held only for the duration of a single request. TLS 1.3. No payload retained on the GPU.
- Whole case files in bulk
- Debtor / creditor PII as a dataset
- Anything used for model training
Chat with
your case archive.
It feels like ChatGPT for your office, but it answers from your files and cites where it found the answer.
Have we handled a similar case before?
Yes. I found three similar matters and the exact pages that show what happened next.
Summarize this case for tomorrow.
Here is the short brief: parties, amounts, deadlines, key documents, and open questions.
Ask naturally.
Write like you would to a careful assistant: what changed, what should I read first, show me similar cases.
Get cited answers.
Every answer points back to the document, page, or record it used.
Stay in control.
Staff review, edit, accept, or reject generated text before it leaves the system.
"Why not just
use ChatGPT?"
Your prompt becomes someone else's tool training corpus.
It can route confidential prompts into provider-controlled infrastructure, retention rules, and model-improvement policies that were not designed around a Polish bailiff office. It does not know your archive, your sygn. akt conventions, or the register of Polish enforcement work. Useful for general questions; the wrong tool for case files.
Your archive is not uploaded to someone else's cloud.
The EU GPU sees only the minimum payload for a single answer or OCR request, retains no payload, and is forbidden — contractually and architecturally — from training on what passes through it. The only model that ever "learns" your archive is the one inside your walls.
What gets installed.
Who does what.
How long it takes.
About one week, end-to-end, for the first office deployment. Your IT team does about two hours of work; we do the rest through an outbound-only operations path. Enterprise is quoted separately when inference must stay on-prem.
Discovery · 30 min
We map your archive structure. You map your concerns. No NDA is needed for the first call; mutual NDA before office-specific details.
Edge server arrives
One 1U appliance, sealed and inventoried, included in the subscription. Your IT team racks it. On-prem inference adds a dedicated server or a validated GPU host.
Index + validation
OCR and embedding jobs run on the edge server. Data stays in place. We validate retrieval quality against representative questions before expanding usage.
Done. Your office works faster.
Similar cases, full-case briefs, and cited answers are ready in minutes instead of buried in folders.
About two hours, end-to-end.
Rack the appliance. Power. Network port. Allow the outbound operations and inference path. On-prem inference uses a local path.
The rest, remotely.
Configuration, ingest, audit, first retrieval validation, user training, and two weeks of support. Recorded, with your consent, for your records.
Priority support on premium.
A spare-appliance plan and recovery runbook are agreed before go-live. Your data stays on the edge either way.
Lexindex Enterprise.
Optional on-prem inference for offices where generative requests cannot leave the building.
Pricing
| Shape | What's included | |
|---|---|---|
| Monthly subscription | Unlimited monthly usage | 3-month discounted trial · edge appliance · provisioning · first index build · software updates · model updates · EU inference · standard support · no query meter |
| Enterprise | Full on-prem inference | Dedicated inference server in your office · local GPU validation · inference runbook · no generative request ever leaves your premises |
The questions you're
already drafting.
Plain answers, written in a buyer's language. If yours is missing, we will answer it on the demo call — and write it into a future version of this section.
What happens if the EU GPU server is unreachable? +
How do you keep one office separate from another? +
Can inference run on our premises? +
Who at Lexindex can access our data? +
What does cancellation look like? +
How is this priced and billed? +
Can the on-prem appliance run air-gapped? +
Can our DPO inspect the configuration? +
What if the EU GPU provider changes? +
Thirty minutes.
Your archive structure.
Real answers.
We will demo on a synthetic archive, then talk through your archive structure and answer the security question you bring. Mutual NDA before office-specific details — within the same hour, electronically. The first call is with someone who has done this before, not a sales rep reading a script.
Tell us about your office.
Leave your contact details and we'll get back within one business day. The first call is with someone who has done this before — not a sales rep reading a script.